About

I’ve spent my career building shared understanding: between users and complex systems, among the people who build software, and between specifications and code. Now I’m studying how AI can do this for software development itself.

For 17 years at Petrobras I’ve worked in nearly every role of software development: requirements and business analysis, mathematical modeling, architecture, programming, databases, infrastructure, DevOps and technical leadership. Having been on both sides of almost every handover taught me a great deal about shared understanding, and about how processes, governance and regulatory requirements challenge it.

The project that shaped me most is Petrobras’ pipeline logistics planning application. It was a long-standing challenge, and the turning point was getting the requirements right. I started by understanding and documenting them in depth, then designed the data model, architecture and implementation, and led the internal and contracted teams that built it. In production since 2019, it simulates a very large pipeline network almost instantly, shows inputs, consequences and validations side by side, and lets the logistics team plan concurrently on the same scenario.

Research has been part of this path from the start: FAPESP-funded undergraduate research and an MSc in Computer Science at Unicamp, advised by Prof. Arnaldo V. Moura; papers on scheduling, pipeline planning and deep learning (International Transactions in Operational Research, ICAPS, BSB); a patent for an assistive pointing device; and Eclipse Foundation committer status at IBM.

Today I’m focused on AI-assisted specification and documentation. My open-source experiments, Tabularium and its successor Tabularium-Quater, a concept-cell format for living specifications that AI agents can read and maintain, explore the question that drives my research: what is the right unit of specification when the reader is also an AI?